flow-based-connectivity-distribution

name	flow-based-connectivity-distribution
version	v1.0.0
last_updated	2026-04-18T00:00:00.000Z
description	Flow-based probabilistic inference for neural connectivity distribution. Uses normalizing flows to model the distribution of possible brain connectomes, enabling uncertainty quantification in connectivity estimation. Supports downstream tasks: network analysis, disease classification, and intervention planning with principled uncertainty estimates.
category	neuroscience
tags	["connectivity","distributional-inference","normalizing-flows","uncertainty-quantification","brain-network","probabilistic-modeling"]
paper	{"title":"Flow-based Connectivity Distribution Inference","published":"2026-04-16","url":"https://arxiv.org/abs/2604.11761"}
activation	flow-based, connectivity distribution, normalizing flows, uncertainty, probabilistic, brain network

Flow-based Connectivity Distribution Inference

概述

使用标准化流（Normalizing Flows）对脑连接组分布进行建模的概率推断方法。与传统点估计方法不同，该方法提供连接估计的不确定性量化，支持下游任务的可靠决策。

核心创新

将连接组推断从点估计提升为分布推断，使用标准化流学习连接空间上的概率分布。

方法论

标准化流架构

class ConnectivityFlow(nn.Module):
    """使用标准化流建模连接分布"""
    
    def forward(self, base_dist):
        # 从简单分布（如高斯）采样
        z = base_dist.sample()
        # 通过可逆变换映射到连接空间
        for transform in self.transforms:
            z, log_det = transform(z)
        connectivity = z
        return connectivity, log_det.sum()

训练策略

1. 似然最大化：最大化观测数据的边际似然
2. 变分推断：使用流作为灵活的后验近似
3. 条件生成：以协变量（年龄、疾病状态）为条件

不确定性量化

• 连接强度不确定性：每条边的后验分布
• 网络指标不确定性：传播连接到图指标的分布
• 分类不确定性：结合连接不确定性进行疾病分类

应用场景

• 连接组推断：从 fMRI/DTI 数据估计连接及其不确定性
• 疾病分类：结合不确定性的鲁棒分类
• 干预规划：考虑不确定性的最优干预策略

参考文献

@article{flow2026,
    title={Flow-based Connectivity Distribution Inference},
    journal={arXiv preprint arXiv:2604.11761},
    year={2026}
}

---

Generated on 2026-04-18